Screaming in the Cloud - Replay - GCP’s Many Profundities with Miles Ward

Starting point is 00:00:00 The core of the decision-making for me is, in every team, you're looking at what skills do you bring to bear and what problem that you're off to go solve for customers. Do the nuances of these products make easier to solve? Welcome to Screaming in the Cloud. I'm Corey Quinn. I am joined today once again by my friend and yours, Miles Ward, who's the CTO at SADA. However, he is, as I think of him, the closest thing the Google Cloud world has to Corey Quinn. Now, let's be clear, not the music and dancing part. That is Forrest Brazil, but Forrest works at Google Cloud, whereas Miles is a reasonably salty third party. Miles, thank you for coming back and letting me subject you to that introduction.

Starting point is 00:00:53 Corey, I appreciate that introduction. I am happy to provide substantial salt. It is easy as I play brass instruments that produce my spit in high volumes. It's the most disgusting part of any possible introduction. For the folks in the audience. It's the most disgusting part of any possible introduction. For the folks in the audience, I am surrounded by a collection of giant sousaphones, tubas, trombones, baritones, marching baritones, trumpets, and pocket trumpets. So Forrest threw down the gauntlet and was like, I can play a keyboard and sing and look cute at the same time.

Starting point is 00:01:21 And so I decided to fail at all three. We put out a new song just a bit ago that's like us thanking all of our customers and partners covering Kool and the Gang's celebration. And I neither look good, play piano, or smiling or capturing any of the notes. I just play the bass part. That's all I got to do.

Starting point is 00:01:37 Ready to take your AWS Cloud security chops to the next level? Then Sanri Security's virtual summit, Access24, happening on Thursday, September 19th, is where you need to be. Access24 is a free virtual summit where you'll get to hear in-depth keynotes from leading voices at AWS, EY, Global Atlantic Financial Group, and more. Tune in as they share real-world case studies explaining how they successfully navigated complex cloud security challenges,

Starting point is 00:02:10 and stick around to get insights into benchmark data for cloud identities, human and non-human. Register for free at sonri.co slash access24 and gain access to the full day's worth of value-packed sessions. No fluff, just the 411 on tactics and strategies from cloud security leaders to help you meet your own cloud security goals. Once again, that's sonri.co slash access24 to register for Access24 on Thursday, September 19th. That's sonri.co slash access24. See you there. So one thing that I didn't get to talk a lot about, because it's not quite in my universe, for one. And for another, it is during the pre-reinvent, pre-invent, my nonsense thing, run-up, which is Google Cloud Next. And my gag a few years ago is that I'm not saying that Google is more interested in what they're building than what they're shipping, but even their conference

Starting point is 00:03:01 is called Next. But don'tiss. So I didn't really get to spend a lot of attention on the Google Cloud releases that came out this year. But given that SADA is, in fact, the, I believe, largest Google Cloud partner on the internet and thus the world. Partner year, three years in a row back to back, baby. Fantastic. I assume someone's watch got stuck or something, but good work. So you have that bias in the way that I have a bias, which is your business is focused around Google Cloud, the way that mine is focused on AWS, but neither of us is particularly beholden to that given company. I mean, you do have the not getting fired as partner, but that's a bit of a heavy lift. I

Starting point is 00:03:42 don't think I can mouth off well enough to get you there. So we have a position of relative independence. So you were tracking Google Next the same way that I track reInvent. Well, not quite the same way I track reInvent. There are some significant differences. What happened at Cloud Next 2021 that the rest of us should be paying attention to? Sure. I presented 10% of the material at the first reInvent. There were 55 sessions. I did six. And so I have been at cloud events for a really long time. I'm really excited about Google's willingness to dive into demos in a way that I think they have been a little shy about. Kelsey Hightower is the kind of notable, deep exception to that. Historically, he's been ready to dive into the kind of heavy hands-on piece, but...

Starting point is 00:04:29 Wait, those were demos? I was just playing Tetris on stage for the love of it. No, man. He really codes all that stuff up, him and the whole team. Oh, absolutely. I'm sorry. If I ever grow up, I wish to be Kelsey Hightower. You and me both. So he had kind of let the charge. We did a couple of fun little demos while I was there, but they've really gotten a lot further into that and I think are doing a better job of packaging the benefits to not just developers, but also operators and data scientists and the broader roles in the cloud ecosystem from the new features that are being launched. And I think different than the in-person events where there's 20,000, 40,000 people in the

Starting point is 00:05:06 audience paying attention, I think they have to work double hard to capture attention and get engineers to tune in to what's being launched. But if you squint and look close, there are some, I think, very interesting trends that sit in the back of some of the very first launches in what I think are going to be whole veins of launches from Google over the course of the next several years that we are working really hard to track along with and make sure we're extracting maximum value from for our customers. So what was it that they announced that is worth paying attention to? Now, through the cacophony of noise, one announcement that I was tied to next was the announcement that GME Group, I believe, is going to be putting their

Starting point is 00:05:47 futures exchange core trading systems on Google Cloud. At which point, that to me, and I know people are going to yell at me and I don't even slightly care, that is the last nail in the coffin of the idea that, well, Google's going to turn this off in a couple of years. Sorry, no, that is not a thing that's going to happen. Worst case, they might just stop investing it as aggressively as they are now. But even that would be just a clown shoes move that I have a hard time envisioning. Yeah. You're talking now over a dozen, over 10 year, over a billion dollar commitments. So you've got to just really, really hate your stock price if you're going to decide to vaporize that much shareholder value, right? I mean,

Starting point is 00:06:30 we think that in Google, stock price is a material fraction of the recognition of the growth trajectory for cloud, which is now basically just third place behind YouTube. And I think you can do the curve math. It's not like it's going to take long. Right. That would require effectively ejecting Thomas Kurian as the head of Google Cloud and replacing him with the former SVP of bad decisions at Yahoo. Sure. Google has no shyness about continuing to rotate leadership. I was there through three heads of Google Cloud.

Starting point is 00:06:59 So I don't expect that Thomas will be the last, although I think he may well go down in history as having been the best. The level of rotation to the focuses that I think are most critical, getting enterprise customers happy, successful, committed, building macro scale systems in systems that are critical to the core of the business on GCP has grown at an incredible rate under his stewardship. So I think he's doing a great job. He gets a lot of criticism, often from Googlers, when I wind up getting the real talk from them, which is, can you tell me what you really think? Their answer is no. I'm like, okay, next question. Can I go out and buy you eight beers? And then, yes, I can. And the answer that I get pretty commonly is that he's brought too much Oracle into Google. And okay, that sounds like a bad thing

Starting point is 00:07:45 because, you know, Oracle. But let's be clear here. What are you talking about specifically? And what they say distills down to engineers are no longer the end-all be-all of everything at Google Cloud. Engineers don't get to make sales decisions or marketing decisions,

Starting point is 00:08:00 or in some cases, product decisions. And that is not how Google has historically been run, and they don't like the change. I get it, but engineering is not the only hard thing in the world, and it's not the only business area that builds value. Let's be clear on this. So I think that the things that they don't like are, in fact, what Google absolutely needs. I think, one, the man is exceptionally intimidating and intentionally just hyper, hyper attentive to his business. So one of my best employees, Brad Svee, he worked together with me to lay out

Starting point is 00:08:36 what was the book of our whole department, my team of 86 people there. What are we about? What do we do? And I wanted this as a memoriam to teach new hires as got brought in. So this is like 38 pages of detail about our process, our hiring method, our promotional approach, all of it. I showed that to my new boss who had come in at the time and he thought some of the pictures looked good. When we showed it to TK, he read every paragraph. I watched him highlight the paragraphs as he went through, and he read it twice as fast as I can read the thing.

Starting point is 00:09:10 I think he does that to everybody's documents everywhere. So there's a level of just manual rigor that he's brought to the practice that was certainly not there before that. So that alone, it can be intimidating for folks, but I think people that are high performance find that very attractive. Well, from my perspective, he is clearly head and shoulders above Adam Solipski and Scott Guthrie,

Starting point is 00:09:34 the respective heads of AWS and Azure, for one key reason. He is the only one of those three people who follows me on Twitter. And honestly, that is how I evaluate vendors. That's the thing. That's the only one of those three people who follows me on Twitter. And honestly, that is how I evaluate vendors. That's the thing. That's the only measure. Yeah.

Starting point is 00:09:49 I've worked for a long time with Slipsky. And I think that it will be interesting to see whether Adam's approach to capital allocation, where he really, I think, thinks of himself as the manager of thousands of startups, as opposed to a manager of a global business, whether that's a more efficient process for creating value for customers than where I think TK is absolutely trying to build a much more unified, much more singular platform. And a bunch of the launches really speak to that, right? So one of the product announcements that I think is critical is this idea of the global distributed cloud, Google distributed cloud.

Starting point is 00:10:24 We started with Kubernetes and then you layer onto that. Okay, we'll take care of Kubernetes for you. We call that Anthos. We'll build a bunch of structural controls and features into Anthos to make it so that you can really deal with stuff in a globally way. Okay, what does that look like further? How do we get out into edge environments, out into diverse hardware? How do we partner up with everybody to make sure that kind of like comparing Apple's approach to Google's approach, you have an Android ecosystem of Kubernetes providers instead of just one place you can buy an outpost. That's generally the idea of GDC. And I think

Starting point is 00:10:58 that's a spot where you're going to watch Google actually leverage the muscle that it already built in understanding open source dynamics and understanding collaboration between companies as opposed to feeling like it's got to be built here. We've got to sell it here. It's got to have our brand on it. I think that there's a stupendous and extreme story that is still unfolding over at Google Cloud.

Starting point is 00:11:21 Now, reInvent this year, they wound up talking all about how what they were rolling out was a focus on improving primitives. And they're right. I love their managed database service that they launched because it didn't exist. Yeah. We're on a slide, it's primitives, not frameworks. I was like, I think customers want solutions, not frameworks or primitives. What's your plan? Yeah. However, I take a different perspective on all of this, which is that is a terrific spin on the big headline launches all miss the re-invent timeline. And oops, so now we're just going to talk about these other things instead. And that's great.

Starting point is 00:11:55 But then they start talking about industrial IoT and mainframe migrations and the idea of private 5G and running fleets of robots. And it's... Yeah, that's a cool product. Which one? I'm sorry. They're all very different things. Private 5G and running fleets of robots. And it's... Yeah, that's a cool product. Which one? I'm sorry, they're all very different things. Private 5G. Yeah, if someone someday will explain to me

Starting point is 00:12:10 how it differs from Wavelength, but that's neither here nor there. You're right, they're all interesting, but none of them are actually doing the thing that I do, which is build websites. I mean, I'm looking for web services. It kind of says it in the name. And it feels like it's very much broadening into

Starting point is 00:12:25 everything. And it's very difficult for me to identify, and if I have trouble with it, I guarantee you customers do, of which services are for me and which are very much not. In some cases, the only answer to that is to check the pricing. I thought Kendra, their corporate information search thing was for me. Then I was at 7,500 bucks a month to get started with that thing. And that is, I can hire an internal corporate librarian to just go and hunt through our Google Drive. Great. Yeah. So there are, or our Dropbox or our Slack, we have like five different information repositories and this is how corporate nonsense starts. Let me assure you. Yes. We call that luxury SaaS. You must enjoy your dozens of overlapping bills for what Workspace gives you as a single flat rate.

Starting point is 00:13:09 Well, we have to do a lot of this stuff too. Google Drive is great, but we use Dropbox for holding anything that touches our customers' billing information just because, to be clear, I do not distrust Google. But it also seems a little weird to put the confidential billing information for one of their competitors on their thing if a customer were to ask about it. So it's the, like, I don't believe anyone's doing anything nefarious, but let's go ahead and just make sure in this case. Go further, man. Vimeo runs on GCP. You think YouTube doesn't want to look at Vimeo stats? Like, they run everything on GCP. So they have to have arrived at a position of trust somehow. Oh, I know how. It's called encryption. You've heard of encryption before. It's the best.

Starting point is 00:13:51 Oh, yes. I love these rumors that crop up every now and again that Amazon is going to start scanning all of its customer content somehow. It's first, do you have any idea how many compute resources that would take? And two, if they can actually do that and access something you are storing in there against their attestations to the contrary, then that's your story because one of them just makes them look bad. The other one utterly destroys their entire business. I think that that's the one that gets the better clicks. So no, they're not doing that. No, they're not doing that. Another product launch that I thought was super interesting that describes, let's call it second place. The third place will be the one where we get off into the technical deep end. But there's a whole set of coordinated work they're calling Cortex.

Starting point is 00:14:35 So let's imagine you go to a customer. They say, I want to understand what's happening with my business. You go, great. So you use SAP, right? Say you're a big corporate shop and that's your infrastructure of choice. There are a bunch of different options at that layer. When you set up SAP, one of the advantages that something like that has is they have kind of pre-built configurations for roughly your business. But whatever behaviors SAP doesn't do, right? And say data warehousing, advanced analytics, regression, projection, stuff like that. Maybe that's somewhat outside of the core wheelhouse for SAP.

Starting point is 00:15:07 You would expect like, okay, I'll bolt on BigQuery. I'll build that stuff over there. We'll stream the data between the two. Yay, I'm off to the races. But the BigQuery side of the house doesn't have this like bitchin' menu that says, you're a retailer and so you probably want to see these 75 KPIs and you probably want to chew up your SKUs in exactly this way. And here's some presets that make it so that this is operable out of the box. So they are doing the three-way combination, consultancies plus ISVs plus Google products

Starting point is 00:15:37 and doing all the pre-work configuration to go out to a customer and go, I know what you probably just want. Why don't I just give you the whole thing so that it does the stuff that you want? If that's the very first one, this little triangle between SAP and BigQuery and a bunch of consultancies like mine, you have to imagine they go a lot further with that, a lot faster, right? I mean, what does that look like when they do it with Epic,

Starting point is 00:15:58 when they go do it with Go just generally, when they go do it with Apache? I've heard of that software, right? Like there's no reason not to bundle up what the obvious choices are for a bunch of these combinations. The idea of moving up the stack and offering full-on solutions, that's what customers actually want. Well, here's a bunch of things you can do to wind up wiring together to build a solution is cool. Then I'm going to go hire a company who's already done that and is going to sell it to me at a significant markup because I just don't care. I pay way more to WP Engine than I would to just

Starting point is 00:16:25 run WordPress myself on top of AWS or Google Cloud. In fact, it is on Google Cloud. You and me both, man. WP Engine is the best. It's great because I... You're welcome. I designed a bunch of the hosting in the back of that. Oh, yeah. But it's also the... Well, it costs a little bit more that way. Yeah. But guess what's not? This is even more expensive than that bill. It's my time spent doing the care and feeding of this stuff. I like giving money to experts and making it their problem. Yeah, I heard it said best. Lego is an incredible business.

Starting point is 00:16:52 I love their product. And you could build almost any toy with it. And they have not displaced all of their plastic toy makers. Right. Some kids just want to buy a little car. Oh, yeah. You can build anything you want out of Lego bricks, which are great, which absolutely explains why they are a reference AWS customer.

Starting point is 00:17:12 Yeah, they're great. But they didn't beat all other toy companies worldwide and eliminate the rest of that market because they had the better primitive, right? These other solutions are just as valuable, just as interesting, tend to have much bigger markets. Lego is not the largest toy manufacturer in the world. They are not in the top five of toy manufacturers in the world, right?

Starting point is 00:17:31 Like, so chasing that thread and getting all the way down into the spots where I think many of the cloud providers on their own internally had been very uncomfortable. Like, you mean I gotta go all the way to building the stuff that they need for that division inside of that company in that geo in that industry? That's maybe like a little too far afield. I think Google has a natural advantage in its more partner-oriented approach to create these combinations that lower the cost to them and to customers to getting at

Starting point is 00:18:01 that solution quick. So getting into the weeds of Google Next, I suppose, rather than a whole bunch of things that don't seem to apply to anyone except the four or five companies that really could use it, what things did Google release that make the lives of people building, you know, web apps better? This is the one. So I'm at Amazon, hanging out as a part of the team that built up the infrastructure for the Obama campaign in 2012. And there are a bunch of Googlers there. And we are fighting with databases. We are fighting so hard, in fact, with RDS that I think we are the only ones that Raju has ever allowed to SSH into our RDS instances to screw with them.

Starting point is 00:18:42 Until now, with the advent of RDS custom, meaning that you can actually get in as root. So where the hell that lands between RDS and EC2 is ridiculous. I just know that RDS can now run containers. Yeah, I know how many things we did in there that were good for us and how many things we did in there that were bad for us. And I have to imagine this is not a feature that they really ought to let everybody have, myself included. But I will say that what all of the Googlers that I talked to at, you know, at first blush, where I'm the evil Amazon guy in to sort of distract them and make them build a system that, you know, was very reliable and ended up winning an election,

Starting point is 00:19:19 was that they had a better database and they had Spanner and they didn't understand why this whole thing wasn't sitting on Spanner. So we looked and I read the white paper and then I got all drooly and I was like, yes, that is a much better database than everybody else's database. And I don't understand why everybody else isn't on it. Oh, there's that one reason you've heard of it. No other software works with it anywhere in the world, right? It's utterly proprietary to Google. Yes, they were kind. Oh, you want to migrate it off somewhere else or a fraction of it? Great. Step one, redo your data architecture. Yeah. Take all of my software everywhere, rewrite every bit of it. All those commercial applications. Yeah. Forget all those. You got

Starting point is 00:19:51 to rewrite those too, right? It was just, it was very much where Google was eight years ago. So for me, it was immensely meaningful to see the launch at Next where they described what they are building and have now built, we have alpha access to it, a Postgres layer for Spanner. Is that effectively, you have to treat it as Postgres at all times versus multimodal access? You can get in and tickle it like Spanner if you want to tickle it like Spanner.

Starting point is 00:20:17 And in reality, Spanner is ANSI SQL compliant. You're still writing SQL. You just don't have to talk to it like a REST endpoint or a gRPC endpoint or something. So it's similar to Azure's Cosmos DB on some level, except for the part where you can apparently look at other customers' data. Exactly. Yeah. You will not have a sweeping discovery of incredible security violations in the structure of Spanner in that it is the control

Starting point is 00:20:46 system that Google uses to place every ad. And so it does not suck. You can't put a trillion dollar business on top of a database and not have it be safe. That's kind of a thing. The thing that I find is the most interesting area of tech right now is there's been this rise of distributed databases. UgaByte or UjaByte, Planetscale or Planetscale, depending on how you pronounce these things. Yeah, why is G such an adversarial consonant? I don't understand why we've all gotten to this place.

Starting point is 00:21:18 Oh, yeah, but at the same time, so you take a look at all these, and they all are speaking Postgres. It is pretty clear that Postgresquille is the thing that is taking over the world as far as databases go. If I were building something from scratch that used... For folks in the back, that's Postgres

Starting point is 00:21:32 SQL for the rest of us. It's okay. It's going to be all right. Same difference, but yeah, it's the thing that is eating the world. Although, recently, I've got to say, MongoDB is absolutely stepping up in a bunch of really interesting ways. I mean, I think the 4.0 release, I'm the guy who wrote the MongoDB on AWS Best Practices white paper, and I would grab a lot of customers.

Starting point is 00:21:51 They've changed it since then. Step one, do not use DocumentDB. If you want to use Mongo, use Mongo. Yeah, that's right. There were a lot of customers I was on the phone with where Mongo had summarily vaporized their data. And I think they have made huge strides in structural reliability over the course of, you know, especially this 4.0 launch, but the last couple of years for sure. And with all the people they've been hiring from AWS, it's one of those, well, look at this, now who's losing important things from production? Right. So maybe there's only actually five humans who know how to do operations. And we just sort of keep moving around these

Starting point is 00:22:23 different companies. That's sort of my assumption on these things. But Postgres, for those who are not looking to depart from the relational model, is eating the world. Yeah. And there's this like basic emotional thing. My buddy Martin, who set up MySQL and took it public and then promptly got it gobbled up by the Oracle people, like there was a bet there that said, hey, there's going to be a real open database. And then squish, like the man came and got it. And so like, if you're going to be an independent open source software developer, I think you're probably not pushing your pull requests

Starting point is 00:22:56 to our friends at Oracle. That seems weird. So instead, I think Postgres has gobbled up the best minds on that stuff. And it works, it's reliable. It's consistent. And it's functional in all these different sort of reapplications and subdivisions, right? I mean, you have to sort of squint real hard, but down there in the guts of Redshift, that's Postgres, right? Like there's Postgres behind all sorts of stuff. So as an interface

Starting point is 00:23:20 layer, I'm not as interested about how it manages to be successful at bossing around hardware and getting people the zeros and ones that they asked for back in a timely manner. I'm interested in it as a compatibility standard, right? If I have software that says I need to have Postgres under here and then it all will work, that creates this layer of interop that a bunch of other products can use. So folks like PlanetScale and UgoBite can say, no, no, it's cool. We talk Postgres. That'll make it so your application works right. You can bring a SQL alchemy and plug it into this or whatever your interface layer looks

Starting point is 00:23:51 like. That's the spot where if I can trade what is a fairly limited global distribution, global transactional management, literally ridiculously unlimited scalability and zero operations. I can hand all the hard parts of running a database over to somebody else, but I get my layer and my software talks to it. I think that's a huge step forward. Backblaze B2B cloud storage helps you scale applications and deliver services globally. Egress is free up to three times the amount of data you have stored and completely free between S3 compatible Backblaze and leading CDN and compute providers like Fastly, Cloudflare, Vulture, and Coreweave. Visit backblaze.com to learn more. Backblaze, cloud storage built better.

Starting point is 00:24:41 I think that there's a strong movement toward building out on something like this if it works, just because, well, I'm not multi-region today, but I can easily see a world in which I'd want to be. So, great. How do you approach the decision between, once this comes out of alpha, let's be clear. Let's turn this into something that actually ships. And no, Google, that does not mean slapping a beta label on it for five years is the answer here. You actually have to stand behind this thing. But once it goes GA... GA is a good thing. Yeah. How do you decide between using that or PlanetScale or YugaByte? Or Cockroach or SingleStore, right? I mean, there's a zillion of them that sit in this market. I think

Starting point is 00:25:18 the core of the decision-making for me is in every team, you're looking at what skills do you bring to bear and what problem that you're off to go solve for customers. Do the nuances of these products make easier to solve? So I think there are some products that the nature of what you're building isn't all that dependent on one part of the application talking to another one or in an event happening someplace else mattering to an event over here. But some applications that's like utterly critical, like totally, totally necessary.

Starting point is 00:25:51 So we worked with a bunch of like Forex exchange trading desks that literally turn off 12 hours out of the day because they can only keep it consistent in one geographical location right near the main exchanges in New York. So that's a place where I go, would you like to trade all day? And they go, yes, but I can't because databases.

Starting point is 00:26:12 So awesome. Let's call the folks on this banner side. They can solve that problem. I go, would you like to trade all day and rewrite all your software? And they go, no. And I go, oh, what about trade all day, but not rewrite all your software? There we go. Now we've got a solution to that kind of problem. So like we built this crazy game, like totally other end of the ecosystem with the Dragon Ball Z people, hysterical. You're like, you literally play like rock, paper, scissors with your phone. And if you get a rock, I throw a fireball and you get a paper, then I throw a punch and we figure out who wins, but they can play these games like Europe versus Japan, thousands of people on each side, real time, and it works. So let's be clear. I have lobbied a consistent criticism at Google for a while now, which is the Google Cloud global

Starting point is 00:26:57 control plane. So you wind up with things like global service outages from time to time. You wind up with this thing is now broken for everyone everywhere. And that, for a lot of these use cases, is a problem. And I said that AWS's approach to regional isolation is the right way to do it. And I do stand by that assessment, except for the part where it turns out there's a lot of control plane stuff that winds up single-tracking through US East 1, as we learned in the great US East 1 outage of 2021. Yeah, when I see customers move from data center to AWS, what they expect is a higher count of outages that last less time.

Starting point is 00:27:37 That's the trade-off, right? There's going to be more weird, spurious stuff. And maybe, maybe if they're lucky, that outage will be over there at some other region they're not using. I see almost exactly the same promise happening to folks that come from AWS and in particular from Azure over onto GCP, which is there will be probably a higher frequency of outages at a per product level, right? So like sometimes like some weird product takes a screwy sideways where there is structural integer dependence between quite a few products. We've actually published a whole internal structural map of like, you know,

Starting point is 00:28:08 it turns out that Cloud SQL runs on top of GCE, not on GKE. So you can expect if GKE goes sideways, Cloud SQL is probably not going to go sideways. The two aren't dependent on each other. You check the status page and Amazon Free RTOS in a region is having an outage today or something like that. And you're like, oh no, that's terrible. First, let me go look up what the hell that is. And do I, am I using it? Absolutely not. Great. At hyperscalers, well, hyperscale, there are always things that are broken in different ways in different locations.

Starting point is 00:28:35 And if you had a truly accurate status page, it would all be red all the time or varying shades of red, which is not helpful. So I understand the challenge there, but very often it's a partition that you are not exposed to, or the way that you architected things, ideally, means it doesn't really matter. And that is a good thing. So raw outage counts don't solve that. I also maintain that if I were to run in a single region of AWS, or even a single AZ, in all likelihood, I will have a significantly better uptime across the board than I would if I ran it myself. Oh, for sure.

Starting point is 00:29:10 For sure. They're way better at ops than you are. And me. Of course. Ridiculous. And they got that way by learning. I think in 2022, it is unlikely that there's going to be an outage in an AWS availability zone by someone tripping over a power cable, whereas I have actually done that.

Starting point is 00:29:26 So there's a, to be clear, in a data center, not an AWS facility that would not have flown. So there is the better idea of going in that direction. But the things like Route 53's control plane, single tracking through US East 1, if you can't make DNS changes in an outage scenario, you may as well not have a DR plan for most use cases. To be really clear, it was a part of the internal documentation on the AWS side that we would share with customers to be absolutely explicit with them. It's not just that there are mistakes and accidents which we try to limit to AZs, but no, go further, that we may intentionally cause outages to AZs if that's what allows us to keep broader service health higher, right? They are not just a blast

Starting point is 00:30:14 radius because you oops pulled the pin on the grenade. They can actually intentionally step on the off button. And that's different than the way Google operates. They think of each of the AZs and each of the regions and the global system as an always-on, all-the-time environment. And they do not have systems where one gets sort of sacrificed for the benefit of the rest, right? Or they will intentionally plan to take a system offline. There is no planned downtime in the SLA, where the SLAs from my friends at Amazon and Azure are explicit to, if they choose to, they decide to take it offline, they can. And that's, I don't know, I kind of want the contract that has the other thing where you don't get to.

Starting point is 00:30:54 I don't know what the right answer is for a lot of these things. I think multi-cloud is dumb. I think that the idea of having this workload that you're going to seamlessly deploy to two providers in case of an outage, well, guess what? The orchestration between those two providers is going to cause you more outages than you would take just sticking on one. And in most cases, unless you are able to have complete duplication of not just functionality, but capacity between those two, congratulations, you've now just doubled your number of single points of failure. You made the problem actively worse and more expensive. Good job. I wrote an article about this, and I think it's important to differentiate between dumb and terrifyingly, shockingly expensive,

Starting point is 00:31:34 right? So I have a bunch of customers who I would characterize as rich, as like shockingly rich, as producing businesses that have 80 plus percent gross margins. And for them, the costs associated with this stuff are utterly rational and they take on that work and they are seeing benefits or they wouldn't be doing it. Of course. So I think the trajectory in technology, you know, this is a quote from a Google engineer. It's just like, oh, you want to see what the future looks like. Hang out with rich people. I went into houses when I was a little kid that had whole home automation. I couldn't afford them. My mom was cleaning house there, but now my house, I can use my phone to turn on the lights. Like, you know, unless us East one is having a problem. Hey, and then no Roomba for you, right? Like utterly offline. So Roomba's now failed to room

Starting point is 00:32:19 conveniently. My lights are Phillips hue and that's on Google. So that baby works, but it is definitely a spot where the barrier of entry and the level of complexity required is going down over time. And it is definitely a horrible choice for 99% of the companies that are out there right now. But next year, it'll be 98. And the year after that, it'll probably be 97. And if I go inside of Amazon's data centers, there's not one manufacturer of hard drives. There's a bunch.

Starting point is 00:32:48 So that got so easy that now, of course, you use more than one. You got to do it. That's just like sort of a natural thing, right? These technologies, it'll move over time. We just aren't there yet for the vast, vast majority of workloads. I hope that in the future, this stuff becomes easier,

Starting point is 00:33:05 but data transfer fees are going to continue to be a concern. Just right in the face. Especially with the Cambrian explosion of data because the data science folks have successfully convinced the entire industry that there's value in those load balancer logs from 2012.

Starting point is 00:33:21 Okay, great. We're never deleting anything again. But now you've got to replicate all of that stuff because no one has a decent handle on lifecycle management and won't for the foreseeable future. Great. To multiple providers so that you can work on these things, that is incredibly expensive. Yeah. Cool tech from this announcement at Next that I think is very applicable and recognize the level of like utter technical mastery and security mastery to our earlier conversation that something like this requires. The product is called BigQuery Omni.

Starting point is 00:33:50 What Omni allows you to do is go into the Google Cloud console, go to BigQuery, say I want to do analysis on this data that's in S3 or in Azure blob storage. Google will spin up an account on your behalf on Amazon and Azure and run the compute there for you. Bring the result back. So just transfer the answers, not the raw data that you just scanned. And no work on your part, no management, no crapola. So there's like, that's multi-cloud. If I've got, I can do a join between a bunch of rows that are in real BigQuery over on GCP side and rows that are over there in S3. The cross-eyedness of getting something like that to work is mind-blowing.

Starting point is 00:34:30 To give this a little more context, just because people, it's difficult to reason about these things. I can either have data that is in a private subnet in AWS that traverses their horribly prized managed NAT gateways and then goes out to the internet and sent there once for the same cost as I could take that same data and store it in S3 in their standard tier for just shy of six full months. That's a little imbalanced, if we're being direct here.

Starting point is 00:34:57 And then when you add in things like intelligent tiering and archive access classes, that becomes something that there's no contest there. If we're talking about things that are now approaching exabyte scale, that's one of those, yeah, do you want us to pay by credit card? Get serious that you can't at that scale anyway. Invoice billing, or do we just like drive a dump truck full of gold bricks and drop them off in Seattle?

Starting point is 00:35:19 Sure. Same trajectory on the multi-cloud thing. So like a partner of ours, Packet Fabric, you know, if you're a big, big company, you go out and you call Amazon and you buy a hundred gigabit interconnect on, they call theirs direct connect. And then you hook that up to the Google one

Starting point is 00:35:35 that's called dedicated interconnect. And voila, the price goes from 12 cents a gig down to two cents a gig. Everybody's much happier. But Jesus, you pay the upfront for that. You got to set the thing up. It takes days to get deployed. And now you're culpable for the whole pipe

Starting point is 00:35:48 if you don't use it up. Like there are charges that are static over the course of the month. So Packet Fabric just buys one of those and lets you rent the slice of it you need. And I think they've got an incredible product. We're working with them on a whole bunch of different projects.

Starting point is 00:36:01 But I also expect like, there's no reason the cloud providers shouldn't be working hard to vend that kind of solution over time. If 100 gigabit is where it is now, what does it look like when I get to 10 gigabit, when I get to 1 gigabit, when I get to half gigabit? Utility priced that for us so that we get to rational pricing. I think there's a bunch of baked-in business and cost logic that is a part of the pricing system where egress is the source of all of the funding at Amazon for internal networking, right? I don't pay anything

Starting point is 00:36:33 for the switches that connect this machine to that machine in region. It's not like those things are cheap or free. They have to be there. But the funding for that comes from egress. So I think you're going to end up seeing a different model where you'll maybe have different approaches to egress pricing, but you'll be paying like an in-system networking fee. And I think folks will be surprised at how big that fee likely is because of the cost of the level of networking infrastructure that the providers deploy. I don't know if you've gone and tried to buy a 40 port, 40 gig switch anytime recently. It's not like they're those little, you know, blue Netgear ones for 90 bucks. Exactly. It becomes this, I don't know. I keep thinking that's not the right answer, but part of it also is like, well, you know, for things I really need local and don't want to worry about the internet's melting today, I kind of just want

Starting point is 00:37:22 to get a, like some kind of Raspberry Pi shoved under my desk for some reason. Yeah. I think there is a lot where as more and more businesses bet bigger and bigger slices of the farm on this kind of thing, I think it's Jassy's line that you're, you know, the fat, the margin, your business is my opportunity. Like there's a whole ecosystem of partners and competitors that are hunting all of those opportunities. I think that pressure can only be good for customers. Miles, thank you for taking the time to speak with me. If people want to learn more about you, what you're up to, your bad opinions, your ridiculous company, et cetera, where can they find you? Well, it's really easy to spell, SADA.com, S-A-D-A.com. I'm Miles Ward. It's at Miles Ward on Twitter. You don't

Starting point is 00:38:07 have to do too hard of a math. It's miles at SADA.com. If you want to send me an email, it's real straightforward. So eager to reach out, happy to help. We've got a bunch of engineers that like helping people move from Amazon to GCP. So let us know. Excellent. And we'll, of course, put links to this in the show notes because that's how we roll. Yeah. Thanks so much for being so generous with your time. And I look forward to seeing what comes out next year from these various cloud companies. Oh, I know some of them already and they're good. Oh, they're super good.

Starting point is 00:38:40 This is why I don't do predictions because like the stuff that I know about, like, for example, I was aware that Graviton 3 was coming. Sure. And it turns out that if you're, like, guess what's going to come up and you don't name Graviton3, it's like, are you simple? Did you not see that one coming? Or if I don't know it's coming and I make that guess, which is not the hardest thing in the world, someone would think I knew and leaked.

Starting point is 00:38:56 There's no benefit to doing predictions. No, it's very tough. Very tough. Happy to do predictions in private for customers. Absolutely. Thanks again for your time. I appreciate it. Cheers. Miles Ward, CTO at SADA. I'm cloud economist Corey Quinn, and this is Screaming in the Cloud. If you've enjoyed this podcast, please leave a five-star

Starting point is 00:39:16 review on your podcast platform of choice. Whereas if you've hated this podcast, please leave a five-star review in your podcast platform of choice and be very angry in your opinion when you write that obnoxious comment, but then it's going to get lost because it's using MySQL instead of Postgres.

Your Ad Here

Screaming in the Cloud - Replay - GCP’s Many Profundities with Miles Ward

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.